Move congestion controller processing to the pacer thread.

webrtc/modules/pacing/alr_detector.cc

Index: webrtc/modules/pacing/alr_detector.cc
diff --git a/webrtc/modules/pacing/alr_detector.cc b/webrtc/modules/pacing/alr_detector.cc
index b31d4482bba37896bb7b0709d7e796d6e8c6eb53..997fd130985f4d2f6a608367dd7801f78fec932f 100644
--- a/webrtc/modules/pacing/alr_detector.cc
+++ b/webrtc/modules/pacing/alr_detector.cc
@@ -36,6 +36,8 @@ AlrDetector::AlrDetector()
 AlrDetector::~AlrDetector() {}
 
 void AlrDetector::OnBytesSent(size_t bytes_sent, int64_t now_ms) {
+  // TODO(nisse): It's unclear what guarantees there are that this
+  // function isn't called before SetEstimatedBitrate.

[stefan-webrtc, 2017/01/16 15:42:46]

I guess the guarantee is here in form of a DCHECK... :)

[nisse-webrtc, 2017/01/16 16:14:03]

Problem was, this thread reorg made this check crash. And then I fixed it by
adding an if (bytes_sent > 0) around the call.

Looks a bit brittle... So I guess the right thing is to tolerate zero
estimated_bitrate_bps_.

   RTC_DCHECK(estimated_bitrate_bps_);
 
   rate_.Update(bytes_sent, now_ms);